22 May 2012

New Results from John Hoddinott on the Productive Safety Net Programme in Ethiopia, and on Cost Benefit Ratios in Nutrition

John Hoddinott is one of those researchers who is very modest and self effacing, but whose work is quoted by nearly everyone in the poverty, food and nutrition field. He has a habit of delivering cutting edge methods to questions we all want the answers to.

I have just read two of John's most recent papers (although he would be the first to note the co-authors and the fact that he is not even the first author on one of them).

The first paper addresses the effectiveness of Ethiopia's Productive Safety Net Programme and the second paper is an update of the Copenhagen Consensus paper series on the cost-benefit ratios of nutrition interventions.


The Productive Safety Net Programme (PSNP) is one of the largest public works, household asset building, food basket, cash transfer programmes in Africa, with more than 7 million people participating since operations began in 2006. It is targeted at chronically food insecure households (defined as three or more months of unmet food needs per year in each of the last 3 years).

Unlike some of the earlier impact evaluations (this one uses the most recent data from 2010), this one cannot use non-participants as the control group, basically because with the programme running for 6 years, the number of households who are eligible for it but have not received it has diminished. This is a good thing, but it means there is no control group in the sense of households untouched by the programme.

So the authors use the difference between 1 and 5 years of participation as their treatment and control groups. They use propensity score matching create a comparison group for those who received 1 year and those who received 5 years of PSNP. In general propensity score matching involves matching participants and non participants on their propensity score (their estimated probability of being a participant--thus allowing lots of matching variables to be combined in one composite) to create two groups that are similar except for their participation (I'm making it sound more straightforward than it is).

What do they find? I will restrict my description to the largest component of the PSNP, the public works. The authors find that it is only after 5 years of participation in the PSNP that the impacts on food security really kick in (see Figure 4.3).

If the average food gap is around 3-3.5 months per year, 5 years participation in the PSNP public works project reduces the gap by just over one month, or about one third of the food gap.

Is one third a big impact for the resources expended? It's hard to say without seeing the cost data, but the really important finding is that the impact seems to be increasing at an increasing rate the longer participation continues. 

Too often in development donors and governments give up because the short term impacts are underwhelming. So the government and donors deserve credit for taking the medium term view with the programme (and with the evaluation).

The Copenhagen Consensus

The second paper is an expansion and update of benefit cost ratios for direct and indirect nutrition interventions as part of the Copenhagen Consensus series of papers.


  • yield increases and reduced yield variability, are put through the IFPRI IMPACT model to translate into underweight rates (using the Smith-Haddad 2000 elasticities) and they generate benefit:cost ratios of 16-50.
  • interventions that improve access to market information via sms technologies give benefit:cost ratios of 8 (the authors project this ratio to rise as costs of sms technology decrease and the returns improve from the sms technology application to higher value crops)
  • updates on a range of micronutrient interventions: the new estimates give iron supplementation a lower benefit to cost ratio than those from previous Copenhagen Consensus papers (8-24 compared to 82-140)
  • first ever benefit:cost ratios for bundled interventions that reduce infant stunting (relating to the health of mothers, care behaviours, health of infants, quantity and quality of child diet). Under the most conservative assumptions they consider (which are indeed very strict), the benefit:cost ratio of such interventions is 15. If the most conservative assumptions are relaxed the benefit: cost ratio ranges between 23.8 and 138.6. In the country with the largest number of undernourished children in the world, India, these benefit:cost ratios lie between 44 and 138.6.
There are lots of caveats to these estimates, and they will--and should--be put through the mill, but they are probably the best we can do at this stage. 

Interesting things to note

  • the fairly large agriculture benefit:cost (BC) ratios, although my guess is that these are probably the most fragile given the cross-country modelling they are premised on
  • the downgrading of iron supplementation BC ratios
  • the high BC ratios for the bundled interventions (mostly direct interventions)
  • the scope for increases in the BC ratios for sms technology in agriculture (and probably even bigger ratios if they were applied to health and nutrition information asymmetries)
You can see how foundational and useful these papers are.  Keep up the good work John.

No comments: